I've been looking at ARM's ML integration framework and have become a bit overwhelmed by which processor my colleagues and I should go for. I'm hoping for a recommendation for the processor we should go for between the M4 M0+ or others.
We're building a brainwave recognition system which reads the brain waves and classifies them into certain groups many times per second. there are about 300 weights in the network and we're sampling at around 200-500Hz so all up it equates to around 150,000 calculations per second. The network will either be an RNN (most likely, we are hoping for) or a feed forward (less likely if we can help it). In both cases, the network will be pre-trained and it's just the weights we're loading onto the MCU. The battery 280mAh needs to be able to run for about 8hrs so we need an ultra low power module with FPU. We're transmitting this data wireless over Bluetooth LE once processed on the device. We're not processing the data on the external device simply because of the transmission limitations of BLE.
We've had a look at the TI CC3220 http://www.ti.com/product/CC3220/description but are also open to STMicro and NXP since they utilise the ARM processors.
Would love recommendations!